AMD Developer Forums - AMD CodeAnalyst

Can't see any source code in CodeAnalyst

AndrewSmart — 2010-03-06T01:34:50 -05.00

Hi folks,

I've been struggling to get this working for a week or two now. I'm trying to time-based profile my application "Midnight.exe" using CodeAnalyst which is integrated to VS2008 Pro.

When viewing the results, under the "Processes" tab, I can see that Midnight.exe has a result of 5.91 Timer samples.

Drilling into this, it then shows the breakdown of processes within Midnight.exe, of which the .exe itself shows 24.68 Timer Samples.

Drilling further, it shows a list of function names with offsets and the number of timer samples for each. Here the symbol names for the functions are being displayed correctly.

The problem comes when I drill down 1 more level into any of these functions. It displays the actual name of the source file that contains the relevant function, but no source code. The source files is listed with Timer Samples = 0 which doesn't seem right

I've made sure the path is set correctly to my source files in the Options dialog within VS, and various other combinations of settings, but nothing seems to bring the source code up.

When trying the "classic.exe" example that comes with CodeAnalyst, this works fine and I can see the actual source code.

Can anyone suggest what might be wrong, or some other things to check as I've run out of ideas

Thanks for any help

oprofile source package

ehasenle — 2010-02-11T05:30:55 -05.00

The CodeAnalyst package is rather big, has lots of dependencies and takes some time to compile. Personally I am used to use the bare-bones oprofile command line tool.

Is it possible that AMD provides an oprofile source package without CodeAnalyst until the changes have been merged upstream?

profiling C++ function within a called DLL

mimarsh2 — 2010-01-29T09:20:45 -05.00

This is my first time using AMD CodeAnalyst. So far it's been great for profiling pure C++ code.

My work uses NI Labview and the data display frontend on our C++ processing DLLs. Can I run my DLL with Labview and profile just the function that I want to look at?

Is this what the API is for? In the time sampling example, a new process is launched. I would like to stick some API code into my DLL and profile and single function.

Thanks in advance.

FP eceptions on equality assignment from array value?

heavensonearth — 2010-01-26T10:10:57 -05.00

Now I have an unexpected question...CodeAnalyst is reporting FP exceptions here:

float a[2], b;
....
{math stuff to calculate b}
....
a[1] = b;

Is this because array values must be exactly 32 bits and b may be represented at higher precision due to denormalized approximations, so it is being truncated to fit into the array? Or is there another possible explanation? I'm quite confused why an equality assignment would cause a floating point exception.

Toolbar doesn't show in VS 2005

linkrocks — 2010-01-18T05:42:19 -05.00

I had Code Analyst working on my PC in VS 2005 and somehow I hid the toolbar and now I can't find it. It's not under View->Otherwindows or the Toolbars drop down list. I tried uninstalling and reinstalling CA without any success of the toolbar reappearing. When I look under tools->options I see CA and its options for General, Directories, Toolbars and even with all of the option checked I still see nothing. I'm running Win 7, but as I said I had it running before on here, I just don't know how to get it working again.

Setting trace point for CodeAnalyst/Linux

matsbror — 2010-01-05T04:26:57 -05.00

I have a program which has a long initialization phase for which I am not particularly interested in measuring performance. Is there a way that I can make a profile that starts at a particular point in the code. I understand this possibility exist for the pipeline simulation (which only exist for Windows), but is there a way I can do a similar thing in event-based profiling on Linux?

Best regards

Mats Brorsson

IRET instructon generating general protection fault

jas_dit — 2010-01-04T13:06:13 -05.00

I am trying to develop a 32 bit linux kernel which will be capable of running 64 bit applications. For this I am running proessor in long mode.

When I am trying to return from 64-bit kernel mode to compatibility user mode via IRET instruction it is givng GP#0. The values it is trying to restore is valid one and EFLAGS.NT is also clear. I am trying to restore the following contents.

00000000 0000002B - SS

00000000 BFB088B0 - SP

00000000 00000200 - EFLAGS

00000000 00000023 - CS

00000000 08048074 - IP

Current CFLAGS is 00000000 00000002

The error code I am getting is 0.

From programmers guide I am not able to get any reason for which it can generate a general potection fault with error code 0.

can somebody please help.

Don't see symbols from .pdb

ols6000@sbcglobal.net — 2009-12-29T18:40:04 -05.00

I am attempting TBP on a .exe under WinXP64. If I use Express mode, I can launch my app and profile successfully.

What I want to do now is take samples while the pgm is already running. I managed this by using Normal mode and leaving the "Launch" box blank. CA took samples, and displayed the results, but in my app, the symbols are missing ("NO SYMBOL").

The .pdb file is in the same directory as the .exe. I used the Tools/CodeAnalysst options/Directories tab to add the directory of the .pdb (and .exe) to Debug File Search Paths. The source (all C++) is in a different directory, shown in the Additional Source Dir box.

How can I get CA to read my symbols for this kind of sampling?

Have the kernel module and oprofile been merged upstream?

colberts — 2009-12-19T06:44:58 -05.00

Have the kernel module and oprofile been merged upstream?

If not, why?

Also, making the changes available as a diff might make supporting more versions of the linux kernel easier, or even automatic. Why limit the versions of kernel Code analyst can be used with.

There are other things that could be done also, like maintain a git repository following the linux kernel developement, and a automated build to check whether the changes to the oprofile kernel module work with any given vanilla kernel. Then Code Analyst would just work with any kernel version.

C# Timer based profiling - bad source view and other problems

Numsgil — 2009-12-12T17:50:06 -05.00

I just downloaded the latest version of CodeAnalyst today to profile my C# project. It's a nice profiler, but I've run in to some issues. Could these be looked at sometime in the future?

See this session (let me know if I need to include anything else).

Issue one:

1. Open the session.

2. Go to System Graph.

3. Double click on the blue bar next to Azimuth.dll.jit

4. Double click the first large blue bar next to Azimuth.DenseLinearAlgebra.QRDecomposition::Decompose+3237.

5. The source view reports a single line (at line 303) as taking 99.64% of the execution time. As near as I can tell it's combined the timings from the entire inner loop of my QR decomposition function.

6. I thought this was an artifact of running in release with optimizations, but if you go back to the Azimuth.dll.jit graph, and double click on other blue bars, you get correctly taken to lines of code that make sense to me.

So as near as I can tell the Azimuth.dll.jit graph is fine, and gives results I would expect, but the source view is totally broken. Any ideas why? Is this a limitation of .NET code?

Issue two:

I'm using the Visual Studio integration.

1. I go to session settings and set "Start with the profiling paused".

2. I start my session. The profiling starts paused, as I would expect.

3. I get my app to start the QR decomposition. A window appears above Visual Studio (my App) and it gains focus.

4. I try to click on the "pause" button for CodeAnalyst inside Visual Studio. But instead of unpausing the profiler, it just makes Visual Studio gain focus instead without unpausing the profiler.

...

Thanks!

Can someone explain something?

Morantex — 2009-12-03T17:12:36 -05.00

I have seen the following in the Code Analyst breakdown of a small function, each line of machine code is executed the same number of times.

mov rax,[rdx]

mov rax,[rcx+rax+00000150h]

add rax,rcx

retnq

Now the first 'mov' line takes 4 CPU clocks, the second 'mov' line takes 2 CPU clocks, but the 'add' line takes 63 CPU clocks (this is what appears in the source code stats in Code Analyst).

I'm seeing this sort of unexpected disparity in many places as I profile a large API and test programs. Innocent looking machine instructions that appear to take far longer than similar ones nearby.

Is the displayed 'CPU clocks' reliable? (the 'hot' instructions don't seem to change so I guess they are).

This is in optimized x64 C code running under Vista x64 with an Athlon X2 6000+ and 8GB RAM and benchmarked using Visual Studio 2008 with the Code Analyst addin.

The profiling used is simply "Assess Performance".

Many of the 'hot functions' we are sooming in on, often end up having this kind of bizarre cause, isolated little instructions that seem to be consuming lots of cycles.

I'm no guru on the internals of the x64 processors or the timings of the x64 instruction set, but these numbers do look suspicious.

I'd appreciate any insights into what is going on here.

Thanks

Hugh

Hit Count

Morantex — 2009-12-03T11:57:51 -05.00

CodeAnalyst is proving useful to me but I am frustrated by the lack of a hit count feature.

Can you guys consider adding this to a future version?

If I see a line in some DLL of mine has taken 450 cpu cycles, then it helps to know if that line was executed 1 time, 100 times or a 1,000,000 times.

Without that info it is very hard to assess the overall importance of some suspected hot spot.

Thanks

Hugh

the selected session does not have any data

afabri — 2009-12-02T14:38:28 -05.00

After having installed CodeAnalyst on a Turion64 bit, XP64 machine I try to run CodeAnalyst on the example program. It runs the example program and when it is finished a window pops up with the text "the selected session does not have any data". The same happens with my own programs.

Any advice?

andreas

debug vs release mode

afabri — 2009-12-02T04:25:19 -05.00

Hello,

I get started with the CodeAnalyst, and reading the chapter "Preparing an Application for Profiling" I am already confused. I would like to analyze code in release mode, because otherwise no inlining of C++ happens at all, so my idea is to compile in release mode, but add /Zi and /DEBUG.

Is that the right approach?

Best regards,

andreas

Preparing an Application for Profiling

L2 and L3 cache hit/miss measurements

jyost — 2009-10-30T19:49:39 -05.00

I've got a very simple loop that is accessing an array - it's just a very simple test program. I've noticed that the number of read requests to the L3 cache (event 0x4E0) is many times greater than the number of L2 cache misses (event 0x7E). I'm using a unit mask of 0xF1 for event 0x4E0, and a unit mask of 0x6 for event 0x7E. So I'm only looking at data, not instructions. What I'm wodering is: where are all these read requests to L3 coming from, if not from L2 cache misses? The ratio is almost exactly 32 to 1, which is itself kind of suspicious. Is the cache line size somehow involved in a way that is not obvious to me?

memory accesses question

jyost — 2009-10-30T12:53:38 -05.00

I'm kind of new to CodeAnalyst and I find that I frequently get results that I don't expect or have difficulty interpreting.

Here's an example. Consider the following little test program:

#include <stdlib.h>
#include <string.h>

main()
{
    size_t size = 1000000;
    int iters = 1000;

    unsigned char *buf = (unsigned char *)malloc (size);

    register unsigned char sum = 0;

    for (register int i = 0; i < iters; i++)
    {
    for (register int j = 0; j < size; j++)
        sum += buf[j];
    }

    return sum;
}

Compiled as follows:

g++ -o simple simple.cpp

(So - no optimization.)

I would expect this to do ony reads, and no (or very few) writes to main memory. In fact, if I look at events 0x6C (reads) and 0x6D (writes), it seems to do about as many reads as writes, if I'm interpreting the results correctly. Hmmm ... Maybe "sum" isn't being put in a register, in spite of the "register" keyword. That's the only theory I have. But I'm not sure that I believe that.

The actual results I got from one run were 7566 for reads, 31897 for writes and 3128 for DRAM accesses - all with a sample period of 10,000. And ... hmmm ... maybe that sample period should be 500,000. But, still ...

Another question: Why is event 0xE0 (DRAM accesses) not equal to the sum of event 0x6C (reads) and 0x6D (writes)?

What I'm ultimately trying to determine is if a real program (not the above test) is bumping up against memory bandwidth limits, but I'm not sure which event or events I should look at. BTW - I have looked at Paul Drongowski's "Basic Performance Measurements ..." document, which is certainly very helpful, but still leaves me with some questions. (Maybe I'm just thick!)

Perfmon2 on AMD64 running RHEL5.3

aehusain — 2009-10-25T02:02:00 -05.00

Dear All,

I am interested in installing perfmon2 on my AMD 64 box running RHEL5.3 server. I tried installing this using 'yum', but it seems the library only works for itanium 64 archs and not for amd64.

I also tried to get the latest patch from sourceforge but it gives errors while patching the kernel file.

Please suggest me if someone is able to achive this task.

Thanks!

Ata

A couple of instructions take 15000 cycles?

MackTuesday — 2009-10-22T18:31:49 -05.00

The attached code shows the function where I'm getting this behavior. Disassembly lines are interspersed among C++. The offending address is 0x004015C6. If I run either the "Assess Performance" or "Time-based Profile", CodeAnalyst reports this line as taking over 15000 clock cycles. Why is this, and what do I do about it?

codeanalyst from opensuse 11.1 contrib

strieben — 2009-10-20T09:02:06 -05.00

Hello,

I manually installed the codeanalyst package from openSUSE 11.1 Contrib (http://download.opensuse.org/repositories/openSUSE:/Factory:/Contrib/openSUSE_11.1/x86_64/) and ran it as root user. I can go through the tutorial until the step where i would actually run the program. There i receive the GUI error.

Error starting profile: Failed to setup OProfile.

When looking at /var/log/messages, I get the following line:

Oct 20 14:50:21 sedici sudo: root : TTY=unknown ; PWD=/opt/CodeAnalyst/bin ; USER=root ; COMMAND=list

/opt/CodeAnalyst is where bin, lib64, sbin and share are located. I wonder what this 'list' command is all about. Because there is absolutely no such command in my path. I also think that the kernel oprofile module has been successfully loaded, since the 'rccodeanalyst start' script runs with no errors and i get the following, when doing 'ls /dev/oprofile':

0 2 backtrace_depth buffer_size cpu_buffer_size dump pointer_size
1 3 buffer buffer_watershed cpu_type enable stats

Thanks in advance!

Running xperf while CodeAnalyst is installed bluescreens

azverkan@hotmail.com — 2009-10-20T01:38:36 -05.00

Running the command below while CodeAnalyst is installed will result in a BSOD on Vista 64.

xperf -on Latency -stackwalk Profile

Microsoft (R) Windows Debugger Version 6.10.0003.233 AMD64
Copyright (c) Microsoft Corporation. All rights reserved.

Loading Dump File [C:\Windows\MEMORY.DMP]
Kernel Summary Dump File: Only kernel address space is available

Symbol search path is: SRV*c:\symbols*http://msdl.microsoft.com/download/symbols
Executable search path is:
Windows Server 2008/Windows Vista SP1 Kernel Version 6001 (Service Pack 1) MP (4 procs) Free x64
Product: WinNt, suite: TerminalServer SingleUserTS
Built by: 6001.18295.amd64fre.vistasp1_gdr.090721-0152
Machine Name:
Kernel base = 0xfffff800`02662000 PsLoadedModuleList = 0xfffff800`02827db0
Debug session time: Mon Oct 19 22:26:26.272 2009 (GMT-7)
System Uptime: 0 days 0:06:02.378
Loading Kernel Symbols
...............................................................
................................................................
................................................................
.
Loading User Symbols

Loading unloaded module list
............
*******************************************************************************
*                                                                             *
*                        Bugcheck Analysis                                    *
*                                                                             *
*******************************************************************************

Use !analyze -v to get detailed debugging information.

BugCheck 1E, {ffffffffc0000096, fffffa6009c12838, 0, 0}

*** ERROR: Symbol file could not be found. Defaulted to export symbols for AMPCORE.sys -
Probably caused by : AMPCORE.sys ( AMPCORE!AmdPcoreAddRecord+f38 )

Followup: MachineOwner
---------

1: kd> !analyze -v
*******************************************************************************
*                                                                             *
*                        Bugcheck Analysis                                    *
*                                                                             *
*******************************************************************************

KMODE_EXCEPTION_NOT_HANDLED (1e)
This is a very common bugcheck. Usually the exception address pinpoints
the driver/function that caused the problem. Always note this address
as well as the link date of the driver/image that contains this address.
Arguments:
Arg1: ffffffffc0000096, The exception code that was not handled
Arg2: fffffa6009c12838, The address that the exception occurred at
Arg3: 0000000000000000, Parameter 0 of the exception
Arg4: 0000000000000000, Parameter 1 of the exception

Debugging Details:
------------------

EXCEPTION_CODE: (NTSTATUS) 0xc0000096 - {EXCEPTION} Privileged instruction.

FAULTING_IP:
AMPCORE!AmdPcoreAddRecord+f38
fffffa60`09c12838 0f32            rdmsr

EXCEPTION_PARAMETER1: 0000000000000000

EXCEPTION_PARAMETER2: 0000000000000000

DEFAULT_BUCKET_ID: VISTA_DRIVER_FAULT

BUGCHECK_STR: 0x1E

PROCESS_NAME: System

CURRENT_IRQL: 0

LAST_CONTROL_TRANSFER: from fffff80002690e67 to fffff800026b6650

STACK_TEXT:
fffffa60`0afcbff8 fffff800`02690e67 : 00000000`0000001e ffffffff`c0000096 fffffa60`09c12838 00000000`00000000 : nt!KeBugCheckEx
fffffa60`0afcc000 fffff800`026b64a9 : fffffa60`0afcc738 fffffa60`005f6290 fffffa60`0afcc7e0 00000000`00000005 : nt! ?? ::FNODOBFM::`string'+0x29317
fffffa60`0afcc600 fffff800`026b508d : 00000000`00000000 fffff800`026205df fffffa80`04975640 00000000`00001000 : nt!KiExceptionDispatch+0xa9
fffffa60`0afcc7e0 fffffa60`09c12838 : fffffa60`09c0e25b fffffa80`c0010000 fffffa80`04fe8c10 fffffa80`0d082650 : nt!KiGeneralProtectionFault+0xcd
fffffa60`0afcc978 fffffa60`09c0e25b : fffffa80`c0010000 fffffa80`04fe8c10 fffffa80`0d082650 00000000`00000002 : AMPCORE!AmdPcoreAddRecord+0xf38
fffffa60`0afcc980 fffffa60`09c11f6b : 00000000`00000000 00000000`00000000 00000000`00000000 00000000`00000000 : AMPCORE!AmdPcoreClearAllEvents+0x16b
fffffa60`0afcc9d0 fffffa60`09c13030 : fffffa60`0afccc08 fa60005f`00000020 00209b00`00000000 fffffa60`00fa422e : AMPCORE!AmdPcoreAddRecord+0x66b
fffffa60`0afccb80 fffffa60`0afccc08 : fa60005f`00000020 00209b00`00000000 fffffa60`00fa422e 00000000`00000000 : AMPCORE!AmdPcoreAddRecord+0x1730
fffffa60`0afccb88 fa60005f`00000020 : 00209b00`00000000 fffffa60`00fa422e 00000000`00000000 fffffa80`05f75190 : 0xfffffa60`0afccc08
fffffa60`0afccb90 00209b00`00000000 : fffffa60`00fa422e 00000000`00000000 fffffa80`05f75190 fffffa60`0afccc08 : 0xfa60005f`00000020
fffffa60`0afccb98 fffffa60`00fa422e : 00000000`00000000 fffffa80`05f75190 fffffa60`0afccc08 0000fa80`04467890 : 0x209b00`00000000
fffffa60`0afccba0 00000000`00010000 : 00000000`00000000 00000000`00000000 00000000`00000001 fffff800`0291fd50 : ecache!EcDispatchReadWrite+0x8a6
fffffa60`0afcccd0 00000000`00000000 : 00000000`00000000 00000000`00000001 fffff800`0291fd50 fffffa80`0d082650 : 0x10000
fffffa60`0afcccd8 00000000`00000000 : 00000000`00000001 fffff800`0291fd50 fffffa80`0d082650 00000000`00000000 : 0x0
fffffa60`0afccce0 00000000`00000001 : fffff800`0291fd50 fffffa80`0d082650 00000000`00000000 00000000`00000000 : 0x0
fffffa60`0afccce8 fffff800`0291fd50 : fffffa80`0d082650 00000000`00000000 00000000`00000000 00000000`00000000 : 0x1
fffffa60`0afcccf0 fffff800`028dafd3 : 00000000`00000000 fffffa80`04fb1880 00000000`00000080 fffffa80`0d082650 : nt!EtwpLogger+0x1f8
fffffa60`0afccd50 fffff800`026f0816 : fffffa60`005ec180 fffffa80`04fb1880 fffffa60`005f5d40 00000000`00000001 : nt!PspSystemThreadStartup+0x57
fffffa60`0afccd80 00000000`00000000 : 00000000`00000000 00000000`00000000 00000000`00000000 00000000`00000000 : nt!KiStartSystemThread+0x16

STACK_COMMAND: kb

FOLLOWUP_IP:
AMPCORE!AmdPcoreAddRecord+f38
fffffa60`09c12838 0f32            rdmsr

SYMBOL_STACK_INDEX: 4

SYMBOL_NAME: AMPCORE!AmdPcoreAddRecord+f38

FOLLOWUP_NAME: MachineOwner

MODULE_NAME: AMPCORE

IMAGE_NAME: AMPCORE.sys

DEBUG_FLR_IMAGE_TIMESTAMP: 49f88bdc

FAILURE_BUCKET_ID: X64_0x1E_AMPCORE!AmdPcoreAddRecord+f38

BUCKET_ID: X64_0x1E_AMPCORE!AmdPcoreAddRecord+f38

Followup: MachineOwner
---------

About the measurement of L2 Cache on Opteron

yingbo — 2009-10-19T08:36:03 -05.00

I'm evaluating program performance on AMD Opteron 270 with OProfile. I refer to "Basic Performance Measurements for AMD Athlon™ 64, AMD Opteron™ and AMD Phenom™ Processors" by Paul J. Drongowski.

Paul's artcile introduces two methods for L2 cache. One is direct method and the other is indirect method.

Direct method:
L2 request rate = (L2_requests + L2_fill_write) / Ret_instructions
L2 miss ratio = L2_misses / (L2_requests + L2_fill_write)

Indirect method:
IC_misses = IC_refills_L2 + IC_refills_sys
DC_misses = DC_refills_L2 + DC_refills_sys
L2_requests = IC_misses + DC_misses + L2_requests_TLB
L2 request rate = L2_requests / Ret_instructions
L2_misses = IC_refills_sys + DC_refills_sys + L2_misses_TLB
L2 miss ratio = L2_misses / L2_requests

I have some questions about L2 Cache measurement.

1. How to compute L2_request_TLB in the indirect method?
My understanding is L2_request_TLB is equal to the sum of L1_ITLB_MISS_AND_L2_ITLB_MISS and L1_DTLB_AND_L2_DTLB_MISS. Event REQUESTS_TO_L2 has a mask bit (0x4) for TLB. I measured mcf and vortex in SPEC2000.

opcontrol --event=REQUESTS_TO_L2:50003:0x4--event=L1_ITLB_MISS_AND_L2_ITLB_MISS:50003 --event=L1_DTLB_AND_L2_DTLB_MISS:50003 --image=mcf.exe,vortex.exe

L1_DTLB_AND_L2_DTLB_MISS|REQUESTS_TO_L2:0x4|L1_ITLB_MISS_AND_L2_ITLB_MISS:50003|
samples|      %| samples|      %| samples|      %|
-------------------------------------------------------------------------
     1377 100.000      1664 100.000      0 100.000 mcf.exe
     1192 100.000        10 100.000      1816 100.000 vortex.exe

The is a big discrepancy between REQUESTS_TO_L2:0x4 and (L1_ITLB_MISS_AND_L2_ITLB_MISS + L1_DTLB_AND_L2_DTLB_MISS). Which is appropriate?

2. How to compute L2_request?

Direct method: L2_requests + L2_fill_write
Indirect method: IC_misses(IC_refills_L2 + IC_refills_sys) + DC_misses(DC_refills_L2 + DC_refills_sys) + L2_requests_TLB

1) Direct method
opcontrol --event=L2_CACHE_FILL_WRITEBACK:50003 --event=REQUESTS_TO_L2:50003:0x7 --image=mcf.exe, vortex.exe

2) Indirect method
opcontrol --event=DATA_CACHE_REFILLS_FROM_L2_OR_SYSTEM:50003--event=INSTRUCTION_CACHE_REFILLS_FROM_L2:50003 --event=INSTRUCTION_CACHE_REFILLS_FROM_SYSTEM:50003--event=REQUESTS_TO_L2:50003:0x4 --image=mcf.exe,vortex.exe

INSTRUCTION_CACHE_REFILLS_FROM_L2|INSTRUCTION_CACHE_REFILLS_FROM_SYSTEM|REQUESTS_TO_L2:0x4|DATA_CACHE_REFILLS_FROM_L2_OR_SYSTEM |
samples|      %| samples|      %| samples|      %| samples|      %|
------------------------------------------------------------------------
     2251 100.000        15 100.000      2160 100.000      7587 100.000 vortex.exe
     1 100.000           0 100.000      1660 100.000     10491 100.000 mcf.exe

L2_request_mcf_indirect = 2251 + 15 + 2160 + 7587 = 12013
L2_request_vortex_indirect = 1 + 1660 + 10491 = 12152

There is a VERY BIG discrepancy between L2_request computed with direct and indirect methods. Why?

3. Are the following statements right?

1) INSTRUCTION_CACHE_REFILLS_FROM_SYSTEM is equal to L2_CACHE_MISS:0x1.
2) DATA_CACHE_REFILLS_FROM_SYSTEM is equal to L2_CACHE_MISS:0x2.

Any suggestion is welcome! Appropriate measurement parameters are very necessary and important. We should have a unified version

Step 0 problem - error when drilling down to source code, even for classic.exe

fdc2005 — 2009-10-08T00:41:10 -05.00

I have just installed CodeAnalyst 2.94.718.0, and when I run the classic.exe example that comes with the code, I get an error (details below) when I try to drill down into the source-code.

Help | System Information included at bottom.

Here is what I am doing:

- Start | CodeAnalyst.exe

- File | new | project name = classic | OK

- On the "session settings dialog":

1) Changed to "time-based profile"

2) Launch= "C:\Program Files\AMD\CodeAnalyst\SampleCode\classic\Release\classic.exe"

3) Enable "stop data collection when the app exits" and "profile the duration of the app execution"

- Click OK

- Click Green "start profiling" icon.

- Classic.exe runs, finishes in about 10-15 seconds.

- CodeAnalyst loads SystemData/SystemGraph/Processes tabs.

- When I doubleclick on "classic.exe" from the Module name list on the System Data tab, there is a few second pause, and then I get a dialog title "Code Analyst Error". The error is "The specified module (\\Device\HarddiskDmVolumes\PhysicalDmVolumes\BlockVolume1\Program Files\AMD\CodeAnalyst\SampleCode\classic\Release\classic.exe) cannot be found. No module information can be shown.

I saw another post about how this is a problem if the drive is encrypted or removable, but that is not the case here -- C: is not encrypted, nor is it removable. I tried moving the contents of classic to c:\temp\classic and running it from there, but same result - get a very similar error message.

Can you help?

thanks!

-Frank.

Output from system information (computer name blanked out):

---------------------------------------------------------------------------------

Vendor: AMD, Family: AMD Athlon™ 64 Processor

Family 15 Model 13 Stepping 1

Approximate core frequency is 2211 MHz

Onboard local APIC detected.

2 Processors Installed

Computer Name:xxxxx

Windows Server 2003 Version 5.2 Build 3790

Service Pack 2

Memory:

Percentage of Memory is in use: 75

Total physical memory: 2097151 KB

Free physical memory: 1028396 KB

Total size of paging file : 4194303 KB

Free size of paging file: 3143216 KB

Total virtual memory: 2097024 KB

Free virtual memory: 1952280 KB

Screen display: 1680 X 1050 pixels

Output from "used dll versions" on the help | about dialog:

--------------------------------------------------------------------------

caprof.sys: 3.0.2

dbghelp.dll: 6.7.5

psapi.dll: 5.2.3790

CACommon.dll: 1.2.0

CAK86Disasm.dll: 1.4.10

CAWinTaskInfo.dll: 1.3.5

CAProfEngine.dll: 0.0.0

CASimEngine.dll: 0.0.0

caWindebug.dll: 0.0.0

WinSymbolEngine.dll: 1.3.7

DCConfig.dll: 1.0.0

ViewConfig.dll: 1.0.0

QtGui4.dll: 4.3.0

Exporting CodeAnalyst data from the command line

MeMo_oMeM — 2009-10-07T11:46:05 -05.00

Hi,

There are hundreds (maybe thousands) of objects in the code I am analyzing, so I would like to automate the profiling process. For this purpose, I wrote a script that parses the CSV files exported by the CodeAnalyst GUI to derive the information I am looking for. However this script is useless unless I can find a way to export CodeAnalyst data that does not involve clicking on object names one by one to export them.

I can run the profiler from the command line (using opcontrol) but I could not find a way to generate CSV from the command line, I eventually have to import it into the GUI. The generated data, which is dumped under /var/lib/oprofile, is not human readable.

My question is, is there a way to generate a CSV for each different object from the command line? It would be very useful if the GUI provided a way to export data for *all* objects in separate CSV's, instead of clicking on them one by one to export, such as "export all" (?)

Please tell me if my question is not clear. I will try to clarify with an example if needed. Thank you very much in advance!

NO SYMBOLS

dody — 2009-09-15T15:05:05 -05.00

Hello,

I'm a total beginner with CodeAnalyst and tried to create in Eclipse CDT on Windows XP a HelloWorld application by using the Mingw compiler and the Eclipse C++ HelloWorld template.

I compiled a "debug" executable and run CodeAnalyst on it.

The compiler settings are "g3" for maximum symbol information and optimization was switched of.

There wre results produced but when doubleclicking on this process I get a new window with adresses and behind the message "NO SYMBOL".

Doubleclicking on such a line just opens asembler code. How can I get the code displayed in CodeAnalyst like shown in the tutorials? What is my mistake?

Thank you for your help

module merge view for CodeAnalyst?

vukicevic — 2009-09-14T18:25:21 -05.00

One of the UI pieces that CodeAnalyst seems to share with Intel's VTune is its insistance on separating profiling data per module. I can focus on a process and get a list of modules that are contributing to its time, but often what I really want is an overall merged view of all symbols in those modules for the given process. This is especially true if my code involves multiple modules that are often calling eachother; analyzing problems that involve both (where the time spent is roughly evenly split between the two) is much more difficult if I need to view per-symbol data for each module independently.

Is there a way to obtain a merged symbol view for multiple modules? (That is, if I have both A.dll and B.dll contributing time, I want to view a list of symbols in either module, sorted by number of samples, instead of only being able to view the symbols in A.dll /or/ the symbols in B.dll.)

Loading PDB symbols files from MS

AshMcConnell — 2009-09-08T16:11:00 -05.00

Hi Folks,

I am running Code Analyst through Visual Studio 2008, I am trying to see the symbols from d3d9.dll which i've downloaded / cached locally using MS's PDBs, but they don't seem to be being picked up by Code Analyst. Is there any way to specify extra PDB files?

Thanks for your help!

All the best,
Ash

Error: Unable to open the events file

Diggsey — 2009-08-29T09:55:40 -05.00

I've just installed CodeAnalyst, but every time I try to use it, once the sampling has finished it displays the error message 'Unable to open the events file'

The filename in the error message is the path to the CodeAnalyst install directory plus an assortment of seemingly random chinese characters.

Help me interpret Code Analyst Profile Report

prako — 2009-08-27T06:32:27 -05.00

Hi all,
I profiled an application using Code Analyst. I had made some changes to a function using OpenMP. After Profiling i found the function name appear twice in the report lik:

FunctionName$omp1FunctionName

and

FunctionName.

So now wat wud be the total time taken or rather the total samples for that function? Is it the sum of the samples for both instances or something els? Plz enlighten me.

Thanks in Advance.

Performance Counter selection!

aehusain — 2009-08-26T15:16:00 -05.00

Dear All,

I am trying to select performance counters to measure CACHE, DISK and Memory utilization in AMD Opteron systems.

Can I expect to get to get usage using following performance event monitoring:

1. Cache: Data Cache Accesses and Instruction Cache Acceses

2. Memory: Dram Accesses (with unit mask 0x01 for Page hit)

3. Disk: There are two components: dma and non dma

a. Dram Accesses with unit mask 0x02 for Page Miss

b. hypertransport link0 with unit mask 0x02 for dma'ed data

If anybody has done any of these measurement, please comment if there is some gap in my understanding.

I tried to run a sample copy script (copying a large image 1 Gb in loop) and analyzing the above events, I was not able to conclude much from it.

Thanks for your suggestions!

Regards,

Ata

How to relate execution time to the CPU_CLK_UNHALTED sampling?

yingbo — 2009-08-20T22:29:16 -05.00

Hi @ll,

I used OProfile on AMD Opteron 2.0GHz.

opcontrol --event=CPU_CLK_UNHALTED:3001 --image=gzip.exe

The sampling result is 902,645. The execution time of gzip.exe is 9.281s (linux time command).

Estimated execution time: sampling result * sampling interval / CPU frequency
902,645 * 3001 / 2,000,000,000 = 1.35s. But there is a big discrepancy between the two times.

The event CPU_CLK_UNHALTED means "CPU Clocks Not Halted". Does Oprofile count the latency time caused by cache misses or I/O? These events can cause CPU halt. If not, the CPU_CLK_UNHALTED sampling result makes little sense and cannot represent program performance.

But the document "Basic Performance Measurements for AMD Athlon™ 64,AMD Opteron™ and AMD Phenom™ Processors" says "IPC = Ret_instructions / CPU_clocks", which means CPU_CLK_UNHALTED counts cache misses and I/O waiting time. Is it right?

BTW, when I changed event count to 1,0001 the sampling result is 752,647. And changed to 10,0001, the result is 167,612. Why does not the sampling result scale with the event count?

I'm confused...

Any suggestion is welcome.

Selected Session does not contain any data

RThaden — 2009-08-01T07:20:24 -05.00

Hi,

i have some trouble with the CodeAnalyst 2.94.718.0 on Windows XP with latest SP and a Intel Centrino Platform.

First, I selected the default time based sampling configuration and started my app to be profiled. It runs for 20 s, then says "No session data" "The selected session does not contain any data".

I, then, tried to profile the included classic.exe and the same thing happens no matter if I use the precompiled classic.exe or the one compiled by Visual C++ 2005 express.

Did I miss something? Shouldn't that work right out of the box?

Best regards

Rainer

Bluescreens in ampcore.sys on Win7-64 RC

bright — 2009-07-29T15:01:18 -05.00

Hi,

I installed the most recent CodeAnalyst on Windows 7 64 RC. Lately I'm getting regular bluescreens in ampcore.sys, but only when the machine has been idling for some time (eg. when running overnight). When I get back it's usually bluescreened:

"The bugcheck was: 0x0000001e (0xffffffffc0000096, 0xfffff8800b4ef838, 0x0000000000000000, 0x0000000000000000)."

Is anybody else seeing this?

It happens even if I haven't used CodeAnalyst at all during the session. It might be a bug related to automatic standby/hibernation (although they don't actually seem to work for me on Win7, but manual ones do without this problem)... or maybe it's just a leak that breaks over time?

Diasble and Enable Cores

prako — 2009-07-22T02:11:45 -05.00

Hi All,
I want to run my application on ma system with different number of cores enabled and benchmark it. I've a 4-core AMD Opteron. I wanna run it on a single core, then on a two core and finally with all the four cores enabled..... Thx

Hardware Benchmarking tools - Confusing results on Code Analyst

prako — 2009-07-20T03:15:20 -05.00

Hi,
Im using AMD opteron 4-core PC. When i parallelise(Using OpenMP) and run a serial legacy application on the 4-core i get very inconsistent results when compared to the results on a single or dual core. I want to benchmark my PC to know how well the 4-cores are being utilised and to check the load on each core. I also want to know if the program is bein run on four threads as expected(i've used default settings of openmp to spawn four threads). Could someone please suggest some good tools that would give the above information as well as the information on the memory usage etc....

I've used Code analyst but the results are not clear...
It shows four threads running on a single core.. Its pretty confusing

Thanks

CodeAnalyst failed to start profiling.

MaxEd — 2009-07-19T03:50:11 -05.00

Both CodeAnalyst 2.80 and 2.94 stopped working on my computer some time ago (unfortunately, I don't know exact moment). When I try to start profiling session, I get the following error:

---------------------------
The device driver failed to start profiling, please try again.
Failed (0x80004005) to start the task information capture: 0xffffffff8000ffff
---------------------------

A trial version of Compuware BoundsChecker also hangs when trying to start program, so I think it may be something that breaks all advanced profilers (VerySleepy still works, though).

Can anyone advice me what to do? I think it may be a problem with anti-virus software (though turning protection off on NOD32 did not help) or with some system service, but I'm not sure what to try.

System specs:

CPU: Athlon 64 X2 4200+

RAM: 2Gb

GPU: GeForce 9800 GTX+

OS: Windows XP SP3

Compile Issue - Ubuntu 8.10

ecologist09 — 2009-07-10T11:37:09 -05.00

Hello all,

I have been trying to compile CodeAnalyst (2.8.54) on an Ubuntu 8.10 server. I have worked through many issues but i am stuck on this one. Can any one offer any advice?

FOUND: DISTRIB_ID=Ubuntu DISTRIB_RELEASE=8.10 DISTRIB_CODENAME=intrepid DISTRIB_DESCRIPTION="Ubuntu 8.10"
rm -f *.o oprofile.ko .*.cmd .*.flags *.mod.c
make -C /lib/modules/2.6.27-14-server/build SUBDIRS=/home/aaron/CodeAnalyst-Linux-2.8.54/src/cakm/kernel2.6.27 modules
make[1]: Entering directory `/usr/src/linux-headers-2.6.27-14-server'
Makefile:528: /usr/src/linux-headers-2.6.27-14-server/arch/unknown/Makefile: No such file or directory
make[1]: *** No rule to make target `/usr/src/linux-headers-2.6.27-14-server/arch/unknown/Makefile'. Stop.
make[1]: Leaving directory `/usr/src/linux-headers-2.6.27-14-server'
make: *** [default] Error 2
... Creating backup of oprofile.ko.
`/lib/modules/2.6.27-14-server/kernel/arch/x86/oprofile/oprofile.ko' -> `/lib/modules/2.6.27-14-server/kernel/arch/x86/oprofile/oprofile.ko.orig'
su -c "cp oprofile.ko /lib/modules/2.6.27-14-server/kernel/arch/x86/oprofile && /sbin/depmod -a"
cp: cannot stat `oprofile.ko': No such file or directory
make: *** [install] Error 1
ERROR: Cannot build kernel driver for kernel version "2.6.27-14-server".
Please contact AMD for more information

# uname -r
2.6.27-14-server

Simulate local array using shared memory , but no performance improvement?

codeboycjy — 2009-06-17T04:41:16 -05.00

Currently i'm building kd-tree on Brook+, here is the problem i've encountered now.
i need to pick the edge with the longest length to split in the median.

Here is the brute force way of doing it:

if( splitAxis == 0 )
{
   if( v1.x < splitPosition ) { ... }
   if( v2.x < splitPosition ) { ... }
   if( v3.x < splitPosition ) { ... }
}else if( splitAxis == 1 )
{
   if( v1.y < splitPosition ) { ... }
   if( v2.y < splitPosition ) { ... }
   if( v3.y < splitPosition ) { ... }
}else if( splitAxis == 2 )
{
   if( v1.z < splitPosition ) { ... }
   if( v2.z < splitPosition ) { ... }
   if( v3.z < splitPosition ) { ... }
}

I assume there could be much divergency in the above code.

if the float4 could be access this way:
float4 data; data[0] = 1.0f; which is same with:
float4 data; data.x = 1.0f;

The above code could be improved like this
if( v1[splitAxis] < splitPosition ) { ... }
if( v2[splitAxis] < splitPosition ) { ... }
if( v3[splitAxis] < splitPosition ) { ... }

So i simulated the process by an alternative way.

shared float4 lds[256];

lds[ 4 * instanceInGroup().x + 0 ] = float4( v1.x , v2.x , v3.x , 1.0f );
lds[ 4 * instanceInGroup().x + 1 ] = float4( v1.y , v2.y , v3.y , 1.0f );
lds[ 4 * instanceInGroup().x + 2 ] = float4( v1.z , v2.z , v3.z , 1.0f );

if( lds[ 4 * instanceInGroup().x + splitAxis ].x < splitPosition ) { ... }
if( lds[ 4 * instanceInGroup().x + splitAxis ].y < splitPosition ) { ... }
if( lds[ 4 * instanceInGroup().x + splitAxis ].z < splitPosition ) { ... }

In the current code , i thought there could be much more improvement, but when the code is compile on kernel analyzer.]
The performance is worse than the old one... I don't get it. And the bottle neck of the current code is ALU Ops.
But actualy there are three times ALP ops in the old code than the current one. Why there is no performance??

Sorry, i've post the question by mistake

codeboycjy — 2009-06-17T04:26:19 -05.00

Sorry, i've post the question by mistake

codeboycjy — 2009-06-17T04:16:54 -05.00

Fonts in CodeAnalyst dialogs in VS 2008

daocode — 2009-06-10T13:06:50 -05.00

Hello,

I'm trying CodeAnalyst 2.94.718.0. If run as stand-alone (started from the icon) then everything is OK. But when trying to lunch anything from the VS 2008 I can't see any letters in the dialogs. Even, looking into "Tools->Options" and then finding "AMD CodeAnalyst ..." all dialogs there are also unreadable...

Please, advise something...

daocode